Using fieldable spectrometers and chemometric methods to determine RON of gasoline from petrol stations: A comparison of low-field 1H NMR@80 MHz, handheld RAMAN and benchtop NIR. (15th January 2019)

• Record Type: Journal Article
• Authors: Voigt, Melanie; Legner, Robin; Haefner, Simon; Friesen, Anatoli; Wirtz, Alexander; Jaeger, Martin
• Is Part Of: Fuel. Volume 236(2019)
• Page Start: 829
• Page End: 835
• Publication Date: 2019-01-15
• Subjects: Portable spectrometers -- Research octane number -- Low-field NMR -- Handheld Raman -- Multivariate data analysis -- On-site RON determination; Fuel -- Periodicals; Coal -- Periodicals; Coal; Fuel; Periodicals; 662.6
• DOI: 10.1016/j.fuel.2018.09.006
• Languages: English
• ISSNs: 0016-2361

Abstract

Highlights: RON could be determined using handheld Raman and fieldable 1 H NMR spectrometers. SVR yielded better results than PLS. RON were obtained within repeatability and reproducibility limits for 80–90% of the samples. Fieldable spectrometers and SVR models proved suitable for RON determination at gas stations. Abstract: The Research Octane Number (RON) still is the major physical quantity for the characterization of fuels. Spectroscopy and multivariate data analyses have proven themselves alternatives to the traditional CFR motor. Yet, the utilization of handheld or fieldable instruments has been rarely reported rendering the feasibility of fast and simple near-pump RON determination debatable. In this study, the applicability of a handheld Raman and a portable 1 H NMR spectrometer in combination with chemometrics is demonstrated on a laboratory sample training set and compared to NIR spectroscopy. Qualitative classification of a fuel sample is achieved through Principal Component Analysis. The performance of the fieldable spectrometers using Support Vector Regression for RON prediction is found at least equivalent to earlier studies with more sophisticated and expensive instruments. The analytical method and the validated qualitative and quantitative models are then applied to samples from gas stations. The goodness of the method is expressed both in terms of computational residual mean squared errors and the common experimental reproducibility and repeatability limits. Depending on the method 40–50% of the samples are predicted within 0.2 and 80–90% with 0.7 RON.